May 4, 2025 · 1h 32m · lennys-podcast
Inside Devin: The AI engineer that's set to write 50% of its company’s code this year | Scott Wu
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Cognition co-founder and CEO Scott Wu joins Lenny Rachitsky to explore Devin, the world's first autonomous AI software engineer, explaining how asynchronous agent fleets are transforming developers from manual coders into system architects. Scott breaks down Cognition's reinforcement learning foundations, enterprise onboarding workflows, live agent demos, and the relentless execution driving the company's hyper-growth.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Lenny holds 20.4% of the talking time here. How this is scored →
Lenny as informed peer 3.6 Guest teaching 4.2 Guest disagreement 0.4 Lenny pushing back 0.5
speaking balance: gold is Lenny, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:12:10 Rejecting the eng manager framing
Scott firmly rejects Lenny's framing that managing fleets of Devons turns engineers into bureaucratic engineering managers, arguing instead that it represents a higher level of abstraction akin to Python.
Hardest push from Lenny ▶ 1:10:46 Lenny pressing on managerial fatigueLenny challenges Scott on whether managing dozens of AI agents will make engineering tedious by turning the role into review-heavy middle management.
Biggest teaching moment ▶ 1:04:30 Hardware distribution vs AI velocityScott educates Lenny on why AI's market velocity outpaces the PC, mobile, and internet revolutions due to the complete lack of physical hardware distribution bottlenecks.
Lenny holds their own ▶ 57:14 Connecting moats to Cursor CEO thesisLenny synthesizes past guest commentary on AI defensibility and questions Cognition's competitive positioning relative to developer habits.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Lenny as informed peer | Guest teaching | Guest disagreement | Lenny pushing back | Why |
|---|---|---|---|---|---|---|
| Defining Devin: Jagged Intelligence and Enterprise Scale | 3 | 5 | 1 | 1 | Lenny asks Scott to contextualize Devin's evolution in terms of seniority. Scott reframes seniority using the concept of 'jagged intelligence', explaining that while Devin exceeds humans in specific areas, it remains weaker in others. | |
| The Genesis of Cognition: Reinforcement Learning and Hacker Houses | 4 | 6 | 1 | 0 | Lenny prompts the origin story and observes how AI product names differ from corporate names. Scott delivers a deep technical breakdown distinguishing text imitation learning from high-compute reinforcement learning with automated execution loops. | |
| Personifying Devin: Cultivating Asynchronous Human-Agent Teammates | 3 | 4 | 0 | 0 | Lenny questions the unique decision to personify Devin with a human name and personality. Scott explains how anthropomorphizing the agent sets accurate mental models for users to treat it like an asynchronous junior teammate. | |
| Software Engineering's Shift: From Bricklayer to System Architect | 4 | 4 | 0 | 1 | Lenny explores the practical day-to-day workflow shift and presses Scott on internal PR metrics. Scott articulates the transition from manual bricklayer to system architect as Devins handle routine implementation asynchronously. | |
| The Value of CS Fundamentals and Jevons Paradox in Code | 4 | 5 | 1 | 1 | Lenny asks whether future engineers should still learn to code. Scott strongly affirms the need for foundational computer science knowledge to master abstractions, referencing Jevons paradox to show why engineer demand will actually expand. | |
| Managing Fleets of Devons Across Real-World Developer Friction | 4 | 4 | 0 | 1 | Lenny is startled by the ratio of 15 engineers managing multiple Devins in parallel. Scott explains why developer work is bottlenecked by real-world friction like repros and logs rather than raw syntax generation. | |
| Live Demos: Modifying Web Apps, Autonomous Quizzes, and DeepSeek Wiki | 3 | 4 | 0 | 0 | Scott conducts live screen-share demos modifying the Devin web app, generating a dynamic podcast quiz, and indexing the DeepSeek codebase into a Devin Wiki. Lenny engages playfully as the workflow executes in real time. | |
| Sponsor Message: Attio AI-Native CRM | 4 | 4 | 0 | 1 | Following an ad read, Lenny inquires about how Devin handles large, legacy enterprise codebases compared to startups. Scott demonstrates how Devin hierarchicalizes codebases and integrates with Linear tickets. | |
| Product Philosophy: Well-Defined Tasks and Developer Tool Suites | 4 | 4 | 1 | 0 | Scott explains Cognition's product philosophy of giving Devin well-defined tasks rather than ambiguous problems. He notes the tension between classic startup niche advice and building a broader suite of developer intelligence tools. | |
| The Agent Coding Landscape and Multi-Generational Interfaces | 4 | 4 | 0 | 1 | Lenny surveys the competitive agent landscape including OpenAI, Anthropic, Cursor, and Windsurf. Scott envisions multiple upcoming product generations where human developers will interact purely at the product level rather than reviewing source code. | |
| Moats vs. Stickiness: Multiplayer Code Context and Collaboration | 5 | 5 | 1 | 1 | Lenny asks about defensibility and moats, referencing Cursor CEO Michael Truell's perspective. Scott nuances the debate by differentiating traditional moats from workflow stickiness and multiplayer organizational context. | |
| Model IQ Saturation and the Frictionless Distribution of AI | 3 | 6 | 1 | 0 | Scott offers a contrarian take that base intelligence is already largely solved and that the real challenge is teaching models real-world engineering idiosyncrasies. He contrasts AI's zero-friction software distribution with past hardware-constrained revolutions. | |
| Enterprise Playbook: Championing Devin and Scaffolding Junior Engineers | 3 | 3 | 0 | 0 | Lenny asks how enterprise champions successfully deploy Devin across reluctant engineering teams. Scott shares tactical onboarding advice, emphasizing starting with small PRs and virtual machine testing setups. | |
| Agent Orchestration: API Sub-Agents and Multi-Threaded Context | 4 | 4 | 1 | 2 | Lenny raises the concern that engineers will turn into review-heavy managers rather than creators. Scott pushes back, arguing Devin acts as a task-delegation interface rather than administrative management, while discussing API-orchestrated sub-agents. | |
| Relentless Execution: Startup Principles and Extreme Recruiting | 4 | 4 | 0 | 0 | Scott shares counterintuitive startup lessons, highlighting how Cognition relentlessly applied basic principles such as flying to North Carolina to meet an MIT student's parents to recruit him. | |
| Lightning Round: Book Picks, Favorite Hardware, and Emotional Detachment | 3 | 3 | 0 | 0 | During the lightning round, Scott shares his perspective on startup resilience, emphasizing intense focus combined with Buddhist-like emotional detachment from outcomes, which Lenny connects to executive coaching insights. | |
| The Origin of Devin's Name and Concluding Remarks | 2 | 2 | 0 | 0 | Scott explains the origin of Devin's name as a synthesis of developer prototypes ('Dev Steven' and 'Dev Walden') into a universal developer. Lenny wraps up the conversation warmly. |
Statements from this episode (35)
Wu: Devin is used by Fortune 100 companies and public banks
“So we work with companies of all stages and sizes, you know, on the smallest end, it goes to, you know, startups of just one or two people who are using Devon to build out a lot of their kind of like an initial prototype or initial product all the way up to, y…”
Scott Wu: Reinforcement learning represents the next paradigm shift in AI capabilities
“Reinforcement learning was really working and was going to be the next big paradigm shift in capabilities.”
Wu: Automated code execution feedback loops make RL uniquely powerful for coding
“Code has this whole automated feedback loop, right, where you can run the code, and that is the kind of automated feedback that really feeds into the RL, which makes these models so great at coding.”
Wu: AI product experiences will shift from text completion to autonomous agents
“The product experience was going to shift from, you know, what I'll call, like, text completion To agents, basically.”
Wu: Cognition started in November 2023 and launched Devin in March 2024
“We started in November of 20, 23, which was, yeah, just like hackathon mode. We officially made it into a company around the start of 20, 24. And then our initial launch was in March.”
Wu: Running multiple asynchronous Devin instances is the best workflow
“We found that the, you know, the best workflow really as a team building off stuff was to work with multiple Devons and to run them asynchronously and to kick them off and to only jump in basically as you needed to provide feedback or steer the plan or anythin…”
Scott Wu: Programming will become more important as AI gets more powerful
“And so in that lens, I really think that programming is, if anything, is only going to become more and more important as AI gets more powerful.”
Cognition: Devin merges several hundred PRs monthly into company codebase
“Devon merges like several hundred pull requests into production in the Devon code bases every month, you know, which is, I mean, our whole team is only like 15 engineers, and so it's a pretty sizable fraction of all the code that we write.”
Scott Wu: Devin accounts for about 25% of Cognition's PRs
“It's in the neighborhood of a quarter or so of all of our, yeah”
Scott Wu: Devin will author over 50% of Cognition's code by year-end
“I think by the end of this year, we expect it to be more than half.”
Wu: Aspiring engineers should absolutely still learn to code
“And first of all, the question of, you know, whether you should still learn the code, my answer would be absolutely yes. I think to a large extent, you know, computer science, when you take computer science classes and when you learn these fundamentals, sure, …”
Wu: There will be far more software engineers in a few years
“And so if anything, I think there's going to be, I think there's going to be way more product, way more programmers and way more engineers, you know, a few years from now than there are today.”
Wu: 10x engineering speed will yield over 10x more code
“I think in a kind of zero sum view, you might say, well, we're gonna be 10 times faster at software engineering, and so it means that we're gonna need 10 times fewer software engineers, right? But I think in practice, what really is gonna happen is actually, w…”
Wu: Cognition engineers often manage up to five Devins simultaneously
“Most folks on the team are definitely, yeah, is, are often working with up to five Devons at once, I would say.”
Wu: Agentic workflows will unlock 200% to 1,000% software engineering gains
“I think we're going to shift more and more towards this agentic workflow because, you know, that's in some ways it's kind of like that the way to really get to that, you know, 200%, 500%, a thousand percent gains that, that, that, that we'll be getting to with…”
Devin automatically builds internal wiki representations of indexed codebases for human developers
“Devon sets up its whole wiki with all of its internal understanding. And so when Devon indexes the code base, obviously building a representation of the code base and learning it and improving it over time is one of the big things that Devon does. And funnily …”
Wu: LaunchDarkly's Claire Vo Is a Major Devin User
“Claire Vo from, for example, from LaunchDarkly is is a big Devon user and she loves basically going and scoping out tasks or asking data questions or asking, Hey, what's, you know, what's going on?”
Wu: AI agents succeed on well-defined tasks, not open-ended problems
“I think Devon is best when it is working on tasks that are well-defined. You know, it's one way to put it is you want to be giving Devon tasks, not problems, right? And a lot of these things, you know, like what you just saw, which was kind of like a quick fro…”
Wu: Users have used AI agent Devin to order DoorDash
“We have folks who like order their DoorDash with Devin, for example”
Wu: Generative AI challenges traditional startup advice on niche focus
“One of the, I would say the most common pieces of startup advice out there, I would say is, you know, focus on a really niche cohort, you know, do things that don't scale, you know, make, make, make one use case that's really great. And then you grow from ther…”
Wu: Developers will eventually modify products directly instead of viewing code
“I think the one that we'll get to over the course of several years is probably something where you don't even look at the code at all. And you're actually just looking at your own product and you're just able to look and specify and say, Hey, you know, This bu…”
Scott Wu: No hard barriers prevent competitors entering across AI layers
“At a high level, you know, a lot of different folks had different layers of the AI spectrum, you know the foundation labs or the application layer or so on, you know, I don't think there's any kind of like hard barrier that would prevent others from entering.”
Wu: AI coding agents build defensibility by learning codebases over time
“As you use Devin and as your whole team uses Devin, it's the same thing with an engineer, right? If you're joining on day one versus, you know, so you've been at the company for five years, you know, you wrote half the code yourself. You, you've touched every …”
Wu: Base model intelligence is already sufficient for autonomous software engineering
“I think that my hot take here, which I would give is I think the, I think in terms of base intelligence, we're honestly basically already there. And I think a lot of what we see actually, and what we spend our time on is less so, obviously we don't pre-train o…”
Wu: Engineers not using AI to write code are falling behind
“Where it's, as an engineer, you know, if you're not using AI at all to write code, I mean, you're you know, you're falling behind, honestly. And it is a technology that everyone should have and should be using.”
Scott Wu: Treat Devin like a new junior engineer
“I think the biggest thing I would say is It really is just, you know, treat Devon like your new junior engineer.”
Wu: Devin can spin up sub-Devin agents via API to parallelize work
“So, so for what it's worth, Devon can start other Devins through the API, right? And so, so we, we've seen this happen quite a bit of times where naturally if you have some big tasks that you want to do, like, you know, Devon will do this all the time. It'll, …”
Wu: AI agents decouple context from execution threads across parallel exploration
“With humans, the way I almost say it in, in, in, in technical terms, it's like, there's this coupling of a context and a thread. And what I mean by that is, is like, basically each human can only operate, you know, single-threaded on, on the work that they do,…”
Scott Wu: 18 of Cognition's 27 team members are former founders
“Like I think of our 26 or 27 people total on the team, like I think 18 of us have started our own company before this.”
Wu: Cognition went from hackathons in November to first customers in April
“We had a hackathon in November. We had another hackathon in December. We started the company officially in January. We like got the prototypes out to initial users in February. We did a launch in March. We got our first customers in April.”
Wu: Cognition flew to recruit's parents' home to negotiate full-time hire
“We flew to North Carolina went straight from the airport to his parents' house, had dinner with him and his parents. We talked a lot, you know, it's a really, really nice Gujarati family. We kind of gave them some gifts and just talked to them about it and tri…”
Wu: Cognition hand-wrote recruit's rejection letters to competing job offers
“He was talking with a lot of great companies already. He didn't want to burn any bridges. And so we actually worked with him and basically hand wrote all of his rejection responses to each of the other companies and kind of worked with him on it to say like, y…”
Wu: Detaching personal self-worth from startup outcomes increases founder success
“It is very important to be focused and driven and to really kind of maximize your potential. And then at the same time, it's also very important to not let your own personal emotion get tied up in your success or failure, you know? And I think especially with …”
Cognition's Devin was named after internal prototypes Dev Walden and Dev Steven
“Walden made a virtual, you know, developer version of him, which was called Dev Walden, and then Steven made one of him, which is called Dev Steven, and, you know, we had all this, and then we were kind of combining it all into one thing, and we're like, okay,…”